About Fylume
Fylume is an independent AI research and model-development laboratory. We investigate architectures that minimise unnecessary computation, improve modularity and give people greater control over their computing resources.
What we believe
- Artificial intelligence should be accessible.
- Useful AI should not require an enormous datacentre.
- Users should retain control of their hardware and data.
- Bigger models are not automatically better solutions.
- Models should activate only relevant capabilities when possible.
- Intelligence should be modular rather than unnecessarily monolithic.
- Efficient inference matters.
- Smaller systems deserve serious research.
- AI shouldn’t waste memory and computation.
- Local and sovereign AI should be practical.
- AI research should be transparent and reproducible.
- Open-source development should be encouraged wherever appropriate.
- Research should challenge assumptions, not repeat industry marketing.
- Existing hardware should not be dismissed without evidence.
- Compute budgets should be treated as engineering constraints, not excuses.
How we work
We separate what we hope from what we have shown. Hypotheses are labelled as hypotheses. Results are published with the hardware and settings that produced them. And when something does not work, that is a result too.
We are not trying to make artificial intelligence enormous. We are trying to make it intelligent enough that it doesn’t have to be.